Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
Natural Hazards Review
No full textWorking with the American Red Cross and using the FEMA Individuals and Households Program (IHP) valid registrations data set, we explore the relationship between socially vulnerable populations and the damage that occurred to their homes from Hurricane Michael. We then consider how this information can be used to more effectively target response efforts in these areas, including estimates for sheltering needs and where to focus initial on-the-ground damage assessments.Submitted versio
Water Research
No full textDespite the prevalence of nanoplastics (NPs) in natural and engineered water systems and their association with microbial risks, bacterium-phage interactions have been largely overlooked in the context of biofilm formation. Here, we investigated the effects of positively (PS-NH₂) and negatively (PS-COOH) charged polystyrene nanoplastics (PS-NPs) on dual-species biofilms composed of Escherichia coli (λ+) and Pseudomonas aeruginosa. PS-NPs promoted biofilm formation and stability at environmentally relevant concentrations (e.g., 100–1000 ng/L), with PS-NH₂ exhibiting higher influence. The cellular internalization of PS-NPs increased the reactive oxygen species (ROS) levels by 2.18–2.25 folds, triggered prophage λ activation followed by lysis of E. coli (λ+) after exposure to PS-NPs. Transcriptomic analyses revealed that PS-NPs, especially PS-NH₂, activated the SOS response (2.35–2.63-fold), λ phage replication (2.68–3.97-fold), and interspecies quorum sensing (2.24–5.13-fold), which was verified by the proteomic analyses. Therefore, PS-NPs stimulated protective extracellular polymeric substances (EPS) secretion with eDNA content increased to 325.8–433.8 μg/cm2. Enhanced EPS production contributed to improved biofilm mechanical properties (1.46–1.57-fold as measured by atomic force microscopy) and increased resistance to chlorine disinfection. Metagenomic analysis of pipeline biofilm demonstrated that PS-NPs promoted bacterium-phage interactions and enhanced bacterial antiviral defense systems, which stimulated multi-species biofilm formation and enhanced environmental resilience. Overall, our findings provide novel insights into the interplay between nanoplastics and bacterium-phage dynamics, highlighting increased microbial risks associated with waterborne nanoplastics.Accepted versio
BMC Genomics
No full textBackground: Intramuscular fat (IMF), the white adipose tissue deposited between skeletal muscle fibers, is a key determinant of beef quality due to its contribution to meat flavor, juiciness, and tenderness. However, IMF develops later and grows more slowly, compared to other fat depots such as subcutaneous fat (SF) in cattle. The cellular and molecular mechanisms underlying the delayed development and slow growth of IMF remain poorly understood. We hypothesized that later development and slower growth of IMF compared to SF may, in part, arise from the differences in their progenitor cells.
Results: We performed single-cell RNA sequencing (scRNA-seq) on the stromal vascular fractions (SVFs) from IMF and SF of adult Angus crossbred steers as well as the mononuclear cell fractions (MCFs) from skeletal muscles of newborn Angus crossbred bull calves, with each tissue type collected from two animals. A total of 14,802 cells from 6 animals were sequenced. Clustering analysis revealed that these cells comprised ten cell types, including adipose progenitor cells (APCs), muscle satellite cells (MuSCs), myoblasts, smooth muscle cells, and various immune cell populations. The SF-derived SVF from adult cattle harbored a significantly higher proportion of APCs than the IMF-derived SVF. The MCFs from newborn calves did not contain detectable APCs. Subclustering analysis revealed that the APCs comprised six subpopulations (C0–C5), among which C3 and C5 were absent in the IMF-derived SVF while C1 was markedly less abundant in the IMF-derived SVF than in the SF-derived SVF. Gene set variation analysis and pseudotime trajectory analysis showed that C1 and C3 represented more differentiated APCs, with higher expression of genes involved in adipogenesis, such as PPARG, ADAM12, and PPARGC1A, whereas subclusters C0 and C4 represented undifferentiated, uncommitted APCs, with higher expression of genes involved in DNA replication and cell adhesion, compared to the other subclusters. Conclusions Overall, this single-cell transcriptomics study suggests two potential differences in APCs between IMF and SF in adult cattle: (1) IMF contains fewer APCs than SF; (2) APCs in IMF are adipogenically less committed and less differentiated compared to APCs in SF. These differences may partially explain why IMF develops later and grows more slowly than SF in cattle. This study also suggests that, in cattle, intramuscular fat begins to develop postnatally, challenging the widely held belief that it forms during late gestation.Published versio
Tourism Management
No full textThis study examines how generative artificial intelligence (GenAI) adoption announcements affect firm value in the tourism industry, focusing on online travel agencies (OTAs), hotel companies, and major technology firms. Drawing on signaling theory, two-sided market theory, competitive dynamics, and disruption theory, we analyze GenAI-related announcements made between November 2022 and October 2024 using an event study methodology. The findings reveal that market responses vary depending on the source of the announcement and the type of firm affected. While GenAI announcements from OTAs and hotels generate negative spillover effects for other tourism firms, OTAs experience positive responses to their own and hotel announcements. In contrast, announcements from tech firms trigger negative reactions across OTAs and hotels, suggesting concerns about platform dependency and value displacement. These results highlight the strategic complexity of GenAI signaling and its implications for firm positioning in platform-based industries.Accepted versio
HortTechnology
No full textLettuce (Lactuca sativa) is an economically important leafy green widely grown in greenhouses, yet there are limited data on the interaction between lettuce cultivar and air temperature for many cultivars currently marketed for commercial greenhouse production. Even in climate-controlled greenhouses, internal air temperature can exceed general recommendations, leading to bolting, excessive stem elongation, bitter flavors, and reduced yields. This study aimed to generate benchmark yield and morphological data for 20 lettuce cultivars grown hydroponically in a greenhouse during a fall (20 C mean air temperature) and summer (28 C mean air temperature) production cycle, with harvests at 9 (juvenile stage) and 21 (mature stage) days after transplanting (DAT). For both fall and summer, lettuce grew in a common nutrient film technique system with an average pH and EC of 5.6 and 1.5 dS·m21, respectively, with greenhouse air temperature setpoints of 21 C (day) and 18 C (night), and a target average daily light integral of 17 mol·m22·d21. The experiment was set up as a randomized complete block design with two blocks. Depending on the cultivar, air temperature, and harvest time, lettuce yield and top projected canopy area (TPCA) were significantly different. However, regardless of the cultivar or harvest time, yield (kg·m22·year21) was almost always greater at 20 C (fall) than 28 C (summer). Supraoptimal air temperatures increased TPCA at both 9 and 21 DAT. Supraoptimal air temperatures decreased specific leaf area, resulting in thicker leaf lamina. Chlorophyll concentration was more affected by cultivar than harvest date or air temperature, but prolonged supraoptimal air temperatures decreased chlorophyll concentration at 21 DAT. Benchmarking yield and morphology across cultivars and seasons provides a tool for growers to improve crop selection and production strategies, while informing breeding efforts for improved controlled environment performance with regard to plant architecture and leaf morphology for greenhouses using automated harvesting and packaging.Published versio
Towards an Embodied Composition Framework for Organizing Immersive Computational Notebooks
No full textAs immersive technologies evolve, immersive computational notebooks offer new opportunities for interacting with code, data, and outputs. However, scaling these environments remains a challenge, particularly when analysts manually arrange large numbers of cells to maintain both execution logic and visual coherence. To address this, we introduce an embodied composition framework, facilitating organizational processes in the context of immersive computational notebooks. To evaluate the effectiveness of the embodied composition framework, we conducted a controlled user study comparing manual and embodied composition frameworks in an organizational process. The results show that embodied composition frameworks significantly reduced user effort and decreased completion time. However, the design of the triggering mechanism requires further refinement. Our findings highlight the potential of embodied composition frameworks to enhance the scalability of the organizational process in immersive computational notebooks.Published versio
SeqVLA: Sequential Task Execution for Long-Horizon Manipulation with Completion-Aware Vision-Language-Action Model
No full textLong-horizon robotic manipulation tasks require executing multiple interdependent subtasks in strict sequence, where errors in detecting subtask completion can cascade into downstream failures. Existing Vision-Language-Action (VLA) models such as π0 excel at continuous low-level control but lack an internal signal for identifying when a subtask has finished, making them brittle in sequential settings. We propose SeqVLA, a completion-aware extension of π0 that augments the base architecture with a lightweight detection head perceiving whether the current subtask is complete. This dual-head design enables SeqVLA not only to generate manipulation actions but also to autonomously trigger transitions between subtasks. We investigate four finetuning strategies that vary in how the action and detection heads are optimized (joint vs. sequential finetuning) and how pretrained knowledge is preserved (full finetuning vs. frozen backbone). Experiments are performed on two multi-stage tasks: salad packing with seven distinct subtasks and candy packing with four distinct subtasks. Results show that SeqVLA significantly outperforms the baseline π0 and other strong baselines in overall success rate. In particular, joint finetuning with an unfrozen backbone yields the most decisive and statistically reliable completion predictions, eliminating sequence-related failures and enabling robust long-horizon execution. Our results highlight the importance of coupling action generation with subtask-aware detection for scalable sequential manipulation.Submitted versio
Observational Studies of Rare Quasar Outflows: the FeLoBALs
No full textThe absorption spectra of quasar outflows are studied in order to determine their kinematic and energetic properties and how they affect their host galaxy and its surroundings. If an outflow is sufficiently powerful to have an effect, a process known as active galactic nucleus (AGN) feedback, it can deplete the galaxy's gas reservoir required to produce stars, quenching its star formation rate and thus regulating the host galaxy's evolution. There is a growing body of work studying a rare type of broad absorption line (BAL) quasar that is rich in ion{Fe}{ii} absorption features, as well as ones from similar low-ionization species such as ion{Ni}{ii}, ion{Cr}{ii}, and ion{Fe}{iii}, known as FeLoBALs.
By analyzing the spectra of these objects using data from the Ultraviolet Echelle Spectrograph at the Very Large Telescope (VLT/UVES), we can determine several properties of these outflows, including the hydrogen number density , the hydrogen column density , and the hydrogen ionization parameter . These values can in turn be used to calculate the distance of the outflow from its central source , the mass outflow rate , and the kinetic luminosity . We have found that FeLoBALs can cover a wide parameter space of these properties.
In the first object, quasar SDSS J1130+0411, we find an FeLoBAL system with solar masses per year, among the highest in the literature for any FeLoBAL to date. We additionally determine that this outflow has the capacity to contribute significantly to AGN feedback. We also find seven other outflow systems in this objects, including four outflows, two intervening systems, and a subcomponent of the main BAL. In the object SDSS J2107-0620, we find that the distance of the outflow is parsecs, closer to its central source than any other FeLoBAL to date. We also determine that its is several orders of magnitude too low to contribute to AGN feedback.Doctor of PhilosophyBlack holes have been immensely fascinating, if elusive objects ever since they were first theorized to exist in the early twentieth century. Today, we have direct evidence of these objects thanks to facilities like the Laser Interferometer Gravitational-Wave Observatory (LIGO) and the Event Horizon Telescope (EHT). Astronomers believe that almost every galaxy contains at its heart a supermassive black hole (SMBH), whose mass is between tens of millions to billions of times greater than that of our Sun. Early in their lifespan, galaxies actively accumulate gas onto a disk encircling the back hole, producing large amounts of radiation as the spiraling gas produces friction and releases gravitational energy. This process is so violent that these active galactic nuclei (AGN) can be more luminous than the entire host galaxy. This phenomenon is known as a ``quasar", or ``quasi-stellar object".
Many of these quasars also eject gas away from them during this process. These gaseous outflows can absorb some of the light emitted by the quasar, which astronomers can see by analyzing their spectra using telescopes such as the Very Large Telescope (VLT). We use this world class facility, which consists of four 8-meter telescopes and is situated in the Atacama Desert in Chile, for the studies described in this thesis. The absorption in these spectra appears to be shifted to slightly higher wavelengths compared to the quasar's emission lines due to the Doppler effect. Astronomers can use this wavelength difference to distinguish the outflow from its source and determine the speed at which the gas is moving. If an outflow is powerful enough, it can diminish the galaxy's supply of gas. As this gas is essential in producing stars, powerful outflows can contribute to causing a sharp decrease in the rate of star formation in its source galaxy. This process is known as AGN feedback.
In this dissertation, we discuss the analysis of two quasar outflows using data observed with the VLT. We determine several physical quantities related to these outflows, including the distance between them and their AGN, the rate at which mass is flowing outward, and their kinematic power. One of these outflows, in the object SDSS J1130+0411, has one of the largest mass outflow rates of any other such object studied to date. It is also found to have the ability to contribute to AGN feedback. The second outflow, which was discovered in object SDSS J2107-0620, is closer to its central quasar than any other outflow of its kind known to date. Additionally, its power was found to be far too low to contribute to AGN feedback
On the Optimization of Edge Server Scaling and Placement for Open Radio Access Network (O-RAN) Slicing
No full textManaging edge server resources for O-RAN slicing can be viewed from two complementary perspectives: service providers, who aim to satisfy use-case requirements with minimal leased computing resources, and infrastructure providers, who seek to deploy the minimum number of edge servers across geographically distributed regions.
From the service-provider perspective, this dissertation investigates static scaling of edge computing resources for O-RAN slicing workloads with stringent delay requirements. While static scaling enables proactive resource reservation, it inherently introduces over-provisioning that must be carefully minimized.
To address this challenge, we develop a chance-constrained static scaling framework that accounts for workload uncertainty while guaranteeing processing time requirements. Workload variability is modeled using LDPC decoding, where iteration counts capture stochastic fluctuations induced by wireless channel conditions. Containers are modeled as bins that allocate computing resources under uncertainty, leading to a Stochastic Bin Packing Problem (SBPP) that integrates LDPC asymptotic analysis and the Roofline performance model. Results show that the stochastic formulation achieves higher reliability than deterministic designs, at the cost of moderate additional resource reservation.
From the infrastructure-provider perspective, this dissertation determines the minimum number of edge servers required to support O-RAN slicing in both urban and rural regions under uncertainty. In urban environments, uncertainty arises from dynamic slicing traffic and user activity, modeled using point Poisson processes. In rural, resource-constrained regions, uncertainty is driven by wireless fronthaul behavior, modeled through a random number of retransmissions caused by power-limited communication links.
We formulate the edge server placement problem as a chance-constrained mixed-integer linear program and develop two solution approaches: one based on sample-average approximation and another based on inverse cumulative distribution function (ICDF) transformations. Using realistic wireless testbed measurements, we show that under stable rural channel conditions with low retransmission variability, deterministic designs based on empirical averages closely match stochastic solutions. However, the framework explicitly characterizes the regimes in which uncertainty becomes impactful, providing a principled foundation for data-driven, reliability-aware edge server deployment.Doctor of PhilosophyThis dissertation studies how to manage the computing resources used to support O-RAN slicing, a technology that creates on-demand virtual networks for different wireless services. The work examines the problem from two perspectives: service providers, who want to meet performance requirements while using minimum computing resources, and infrastructure providers, who want to deploy the minimum number of edge servers needed to support these services across different regions.
From the service-provider perspective, the dissertation examines how to reserve resources for O-RAN slicing workloads that must meet stringent delay requirements. A static scaling framework is developed to decide how much computing power each slice should reserve in advance, even when the workload changes unpredictably. The goal is to avoid delays while keeping over-provisioning low. The workload behavior is modeled using the LDPC decoding process, whose iteration counts reflect natural variation in radio processing demand. The problem is formulated as a Stochastic Bin Packing Problem, and results show that the proposed method meets delay requirements more reliably than a more straightforward deterministic approach, albeit at the cost of additional resources.
From the infrastructure-provider perspective, the dissertation investigates how many edge servers are needed—and where they should be placed—to support O-RAN slicing in both urban and rural areas. Urban regions face uncertainty due to fluctuating user activity, while rural regions face unreliable wireless connections and limited power availability. Two stochastic optimization approaches are used to capture these uncertainties. The aim is to place the minimum number of servers while ensuring that service requirements, such as delay, are satisfied with high probability. The placement problem is formulated as a chance-constrained optimization model and solved using established techniques. Results show that accounting for uncertainty reduces deployment costs compared to deterministic methods
Distributionally Ambiguous Stackelberg Combinatorial Games for Submodular Optimization and Camera View-Frame Placement
No full textThis dissertation develops exact solution methodologies for Stackelberg zero-sum games, which model sequential decision-making between an attacker and a defender. Our work specifically addresses challenging settings where the defender's recourse is a complex com- binatorial optimization problem and the attacker faces uncertainty and distributional am- biguity. We analyze these games through two complementary frameworks. Distributionally Robust Optimization (DRO) framework provides a risk-averse attacker with robust attack- ing strategy, while offering defender insights into the most probable threats. In contrast, Distributionally Risk-Receptive (DRR) frameworks provides high-impact strategy for a risk- receptive attacker, thereby serving as a powerful tool for the defender's vulnerability analysis by exposing the system's most critical weakness.
This dissertation makes three primary contributions, each developing novel decomposition methods based on structural insights. First, we introduce and solve a Stackelberg game, where the defender's problem is the camera view-frame placement problem. We address this setting under a DRO framework to explicitly model uncertainty in attack success, incomplete information, and the adversary's varying levels of risk-appetite. For this setting, we develop a cutting-plane-based algorithm that leverages a key geometric property: an optimal placement under one attack remains a feasible recourse under any other, to derive a new class of valid inequalities. Since our algorithm repeatedly solves the defender's problem, placing p camera view frames to maximize the coverage, we also contribute efficient exact methods for p = 1 and novel heuristics for p ≥ 2, validated through simulation experiments of finding a hidden object.
Second, we solve the game when the defender's objective is maximizing k-submodular func- tion, under both DRO and DRR frameworks. To solve problem, we derive valid inequalities from the diminishing property of k-submodular function, and strengthen them further by imposing an ordering over elements in the defender's solution sets. The optimal values from these dual frameworks offer a confidence interval-like range for the defender's expected out- come, where the DRO solution provides robust attack strategies and the DRR solution iden- tifies critical data vulnerabilities. We demonstrate effectiveness of our frameworks through computational experiments on instances of feature selection and sensor placement problems, using Wisconsin breast cancer data and synthetic data, respectively.
Third, we extend the strategic scope to a three-stage Defender-Attacker-Defender (DAD) model with fortification, where the defender's final recourse is the maximization of a sub- modular function. To solve this game, where the standard attacker-defender interdiction game appears as a subproblem, we derive another class of valid inequalities that are con- structed for an arbitrary fortification strategy by leveraging the diminishing return property of the defender's objective function. Empirical validation on real-world datasets with pre- dictive models (e.g., Support Vector Classifiers, logistic regression) confirms the practical impact of our frameworks.Doctor of PhilosophyThe security of critical systems, where failure can lead to severe consequences, requires strategic decision-making against intelligent adversary (attacker) with directly opposing ob- jectives. Unlike passive or random threats, such adversary anticipates the systems user's (defender's) best response to a potential attack and choose their actions accordingly, cre- ating a structured attack where one side's gain directly corresponds to the other's loss. Therefore, rigorous analytical approaches are essential for the decision-making of both the attacker and the defender. This dissertation models these adversarial interactions with the mathematical framework akin to the Stackelberg zero-sum games, which captures the se- quential nature of decision making between a proactive attacker and a reactive defender. A key challenge arises when game parameters are uncertain and historical data is limited to derive precise probability distribution associated with uncertain parameters. This distri- butional ambiguity complicates the search for optimal strategies for attacker and effective vulnerability assessments for defender.
This research addresses this challenge by developing models that analyze two distinct risk preferences of the attacker towards the distributional ambiguity. The risk-averse approach models a cautious attacker, providing conservative strategies that guarantee a consistent level of impact. In contrast, the risk-receptive approach models an aggressive attacker willing to accept higher variability in attack impact in exchange for the potential of a more damaging attack. For a defender, analyzing both models is critical for a comprehensive vulnerability assessment of system, as the former reveals the most probable threats while the latter exposes the most severe ones. Likewise, these frameworks are equally applicable when planning interdiction actions against an adversary (e.g., an evader or enemy) when the decision maker is protagonist.
First, this dissertation provides analytical tools for the security of telerobotic camera systems used in critical roles like surveillance, search and rescue, and satellite imaging, especially in environments where collecting information is difficult for humans. Applying the risk-averse framework offers a dual advantage: it allows a defender to identify vulnerable cameras (or the vehicles carrying them) that are susceptible to attacks from a rational adversary, while also providing a tool for planning interdiction actions to minimize an enemy's information acquisition. To solve this problem, we present an exact cutting planes based algorithm. Moreover, to ensure the framework's computational practicality, we also develop improved algorithms for the defender's underlying placement subproblem, including a faster exact method and heuristics.
Second, we tackle games where the defender's objective is to maximize a k-submodular func- tion. This function, a generalization of standard submodular function, is widely used to model practical optimization problems such as multi-topic influence maximization and fea- ture selection. We explore this framework across two critical problems: the feature selection interdiction problem, where an adversary attacks data features to degrade machine learn- ing model performance, and the weighted coverage interdiction problem, where an attacker blocks sensor installations to minimize maximum coverage. We present finitely convergent exact algorithms for these games and, through computational experiments, demonstrate the practical utility of our models for both decision-makers. For instance, using the Wisconsin Breast Cancer dataset, our risk-receptive model successfully identifies feature attacks that most significantly degrade the predictive model's accuracy.
Third contribution extends the strategic scope of these games by introducing a mathematical model for proactive defense. This framework moves beyond a purely reactive defensive posture, allowing decision-makers to quantitatively assess the value of investing in fortifying assets before an attack occurs, thus enabling more effective long-term security planning